Meta launches Muse Wonder frontier AI models to challenge industry rivals

Meta's Superintelligence Lab has released Muse Wonder and Muse Wonder Full, marking the company's first entry into the frontier AI category to compete with OpenAI and Anthropic.

The new state-of-the-art models are designed for complex tasks like coding and achieve benchmark scores comparable to GPT-6 Astra and Fable 5.1, according to Meta's internal testing. TechCrunch reported that Muse Wonder will be priced at $4 per million input tokens and $20 per million output tokens via the Meta Model API. The larger Muse Wonder Full will be restricted to Muse Code subscribers and will not be available through the API.

A key feature of the release is "active memory," which Meta describes as the ability for models to retain and synthesize user information over long periods. Research scientist Kavya Kopparapu stated the company focused on ensuring the model remembers important details for months. TechCrunch noted that Muse Wonder scored higher on the MemoryBench benchmark than rivals such as Opus, Astra, or Fable.

The Financial Times reported that CEO Mark Zuckerberg characterized the launch as Meta reaching the AI "frontier" while rejecting calls to slow down development. Following the announcement, Meta’s stock price rose by more than 3% as analysts anticipate new business growth from the technology, according to Feeds.

Both models were trained at the Prometheus data center in central Ohio and will be used to power the Muse agent. While Muse Wonder is available via API, Muse Wonder Full is reserved for specific subscribers, representing a shift in scale from the company's previous lightweight Muse Spark model.

Background

The release follows a restructuring of Meta’s AI efforts by Scale AI founder Alexandr Wang in August of last year. Since that reorganization, Meta has released the lightweight Muse Spark model and Muse agent before moving toward these larger frontier systems.